Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Claude says “You're absolutely right!” about everything

github.com

Post reply on HN
Print view View original

431–440 of 560 posts

Re: Claude says “You're absolutely right!” about everything

#431

Post by lazystar »

NohatCoder wrote: post #277

This is such a useful feature. I'm fairly well versed in cryptography. A lot of other people aren't, but they wish they were, so they ask their LLM to make some form of contribution. The result is high level gibberish. When I prod them about the mess, they have to turn to their LLM to deliver a plausibly sounding answer, and that always begins with "You are absolutely right that [thing I mentioned]". So then I don't…

https://news.ycombinator.com/item?id=44860731

well here's a discussion from a few days ago about the problems thia sycophancy causes in leadership roles

Re: Claude says “You're absolutely right!” about everything

#432

Post by furyofantares »

dr_kiszonka wrote: post #335
furyofantares wrote: post #308

Earlier quoted context omitted.

The form of this post is beautiful. "I agree" followed by a completely unrelated reasoning.

They agreed that "this feature" is very useful and explained why.

You're absolutely right.

Re: Claude says “You're absolutely right!” about everything

#433

Post by scotty79 »

catigula wrote: post #390
Applejinx wrote: post #381

Earlier quoted context omitted.

It can't, though. It's language. We don't have a body of work constituting robots talking to each other in words. Hardly fair to ask LLMs not to write like humans when humans constructed everything they're built on.

These models are purposely made to sound more 'friendly' through RLHF

The chat that rejects you because your prompt put it in a bad mood sounds less useful.

Re: Claude says “You're absolutely right!” about everything

#435

Post by cubefox »

9rx wrote: post #376
cubefox wrote: post #365

Earlier quoted context omitted.

Because it's a mirage. People want to be loved, but GPT-4o doesn't love them. It only creates an optical illusion of love.

People want the feelings associated with love. They don't care how they get it. The advantage of "real" love, health wise, is that the other person acts as a moderator. When things start to get out of hand they will back away. Alternatives, like drugs, tend to spiral of out of control when an individual's self-control is the only limiting factor. GPT on the surface seems more like being on the drug end of the spectru…

> People want the feelings associated with love. They don't care how they get it.

Most people want to be loved, not just believe they are. They don't want to be unknowingly deceived. For the same reason they don't want to be unknowingly cheated on. If someone tells them their partner is a cheater, or an unconscious android, they wouldn't be mad about the person who gives them this information, but about their partner.

That's the classic argument against psychological hedonism. See https://en.wikipedia.org/wiki/Experience_machine

Re: Claude says “You're absolutely right!” about everything

#437

Post by wavemode »

hinkley wrote: post #411
lazide wrote: post #292

Earlier quoted context omitted.

You’re likely thinking of calves. Cows (though admittedly ambiguous! But usually adult female bovines) do not drink milk. It’s insidious isn’t it?

If calves aren’t cows then children aren’t humans.

No, you're thinking of the term "cattle". Calves are indeed cattle. But "cow" has a specific definition - it refers to fully-grown female cattle. And the male form is "bull".

Re: Claude says “You're absolutely right!” about everything

#438

Post by visarga »

nojs wrote: post #117

I'm starting to think this is a deeper problem with LLMs that will be hard to solve with stylistic changes. If you ask it to never say "you're absolutely right" and always challenge, then it will dutifully obey, and always challenge - even when you are, in fact, right. What you really want is "challenge me when I'm wrong, and tell me I'm right if I am" - which seems to be a lot harder. As another example, one common…

> I'm starting to think this is a deeper problem with LLMs that will be hard to solve with stylistic changes.

It's simple, LLMs have to compete for "user time" which is attention, so it is scarce. Whatever gets them more user time. Various approaches, it's like an ecosystem.

Re: Claude says “You're absolutely right!” about everything

#439

Post by cubefox »

scotty79 wrote: post #427
cubefox wrote: post #365

Earlier quoted context omitted.

Because it's a mirage. People want to be loved, but GPT-4o doesn't love them. It only creates an optical illusion of love.

If you get a massage from massage machine is it also a mirage? If you use a vibrator is it also a mirage? Why it suddenly becomes an unhealthy mirage if you need words to tickle yourself?

A vibrator still works as intended if you believe it doesn't love you. GPT-4o stops working as intended if you believe it doesn't love you. The latter relies on an illusion, the former doesn't.

(More precisely, a vibrator still relies on an illusion in the evolutionary sense: it doesn't create offspring, so over time phenotypes who like vibrators get replaced by those who don't.)

Re: Claude says “You're absolutely right!” about everything

#440

Post by eurekin »

klik99 wrote: post #338
jjoonathan wrote: post #306

Earlier quoted context omitted.

ChatGPT opened with a "Nope" the other day. I'm so proud of it. https://chatgpt.com/share/6896258f-2cac-800c-b235-c433648bf4...

Is that GPT5? Reddit users are freaking out about losing 4o and AFAICT it's because 5 doesn't stroke their ego as hard as 4o. I feel there are roughly two classes of heavy LLM users - one who use it like a tool, and the other like a therapist. The latter may be a bigger money maker for many LLM companies so I worry GPT5 will be seen as a mistake to them, despite being better for research/agent work.

My very brief interaction with GPT5 is that it's just weird.

"Sure, I'll help you stop flirting with OOMs"

"Thought for 27s Yep-..." (this comes out a lot)

"If you still graze OOM at load"

"how far you can push --max-model-len without more OOM drama"

- all this in a prolonged discussion about CUDA and various llm runners. I've added special user instructions to avoid flowery language, but it gets ignored.

EDIT: it also dragged conversation for hours. I ended up going with latest docs and finally, all issues with CUDA in a joint tabbyApi and exllamav2 project cleared up. It just couldn't find a solution and kept proposing, whatever people wrote in similar issues. It's reasoning capabilities are in my eyes greatly exaggarated.

Post reply on HN